360视频的连续保形投影算法

A Conformal Projection Algorithm for 360 Videos

360视频存储需要经过投影拼合得到平面视频,不同的投影拼合算法对视频的质量影响非常大.为了减少在投影拼合阶段产生的质量损失以及对存储空间的浪费,提出一种连续保形的投影拼合算法,首先通过立体投影将360视频的2个半球面分别投影到一个平面圆上;然后利用Schwarz-Christoffel变换将平面圆保形地映射为正方形;最后将2个正方形连续地拼合在一起,得到360视频对应的平面视频.在实验中,对比了各类球面纹理使用不同投影拼合算法得到的平面图像,分析其拉伸变形的情况并量化不同算法产生的不连续边,结果表明,相比其他主流算法,360视频的连续保形投影算法能够保证投影后图像的拉伸变形较小,同时确保平面视频图像的内部连续性.

A 360 video is stored as a 2D rectangle video through projection and packing.There are many projections and packing layouts,and their efficiency has a great influence on the resulting 360 video quality.In order to improve the quality and reduce the storage waste,we propose a projection algorithm using Peirce quincuncial and conformal mapping.The algorithm projects a hemisphere of the 360 video onto a disk by stereoscopic projection;then uses the Schwarz-Christoffel transformation to conformally map the disk to a square;so does another hemisphere.Then the two squares are stitched together continuously to obtain a 2D rectangle video.The experimental results show that our conformal projection algorithm can generate 360 videos with better continuity and smoother pixel distribution when compared to other commonly used algorithms.